Risk and the multiple: the six levers that decide what the firm is worth at the end

The multiple is argued, not assumed

In the software edition, the exit multiple was a market fact with a growth band around it. Here it is the thesis's largest bet and the critics' main target: services firms "won't trade like software," because the buyer rents labor on the client's terms, hourly billing penalises efficiency, contracts are short, and inefficiency is part of the product.¹ Public comps make the same point without arguing it. The outsourcers that have deployed generative AI widely still trade at low single-digit to low-twenties EV/EBITDA against software peers in the twenties to nineties; one of them has generative AI at more than a thousand customers and an EBITDA margin around ten percent.¹ And the public professional-services index has halved on a multiple basis in a year, with AI concern named as a cause.²

So this chapter does not assume the multiple. It builds the case a buyer would have to accept, metric by metric, and says what the firm is worth if the case is not made.

X1. Professional liability and AI error

What it is. The firm's errors-and-omissions exposure when a platform produces work a client relies on. The engagement terms (what the firm warrants, what it excludes), the insurance (whether the policy covers platform-produced work and what it costs), and the control (D3's review model, which is what an insurer and a client will ask to see). The software edition's AI-governance lever rested on statutes and standards; the services version rests on the professional-liability regime the firm already lives under, which now has to cover a new way of producing the same work.

Line it moves. Risk retained; G&A (premiums, outside counsel).

Typical range and evidence. The hazard is documented in law, where courts have sanctioned filings containing fabricated citations, and the regulator's finding in medical coding is that independent accuracy evidence is scarce.³ No public source gives E&O premium changes for firms converting delivery to AI. Grade: Strong (hazard documented) / Weak (cost unmeasured).

How to run it. Brief the E&O carrier before conversion, with D3's sampling plan in hand; an insurer told after a claim is a different conversation. Put the review model and the AI-disclosure language into the engagement letter at renewal (O5). Keep a claims log by unit type from day one, even when it is empty; a buyer will ask for it.

Kill criteria. Carrier declines to cover platform-produced work, or prices it above the D-group saving for a unit type. Any client claim traceable to a platform-produced unit that passed review (D3's kill, repeated here because it is a liability event).

Metrics. Close: E&O terms and premium; claims history. Year one: premium change; claims by unit type. Year three: the same.

Horizon and stage. Deploy; day 100 to year one.

X2. Licensed sign-off compliance

What it is. P5 found the rules; X2 keeps the firm inside them as delivery changes. Where a licensed or accredited human must perform or sign, the platform prepares and the human signs, and the record shows it. Where an ownership rule forces an alternative structure, the structure is maintained in substance as well as form: a professional body or a state board that finds the platform company directing professional judgment it is not licensed to direct is a kill event for the whole conversion, not a compliance finding.

Line it moves. Risk; the conversion ceiling (P5's reserved share).

Typical range and evidence. Statutes and professional rules, as P5. Governance attention is rising: the international ethics standard-setter for accountants has opened a workstream on private-equity investment in accounting firms.⁴ Grade: Strong (rules); cite each from its text.

How to run it. Design the sign-off step into D1's unit flow for every reserved unit, so the licensed human's review is recorded, not assumed. Keep the alternative structure's governance real: separate boards, separate decisions on professional matters, minutes that show it.

Kill criteria. Any regulatory finding on the conversion.

Metrics. Year one and three: zero findings; share of reserved units with recorded sign-off at 100 percent.

Horizon and stage. Deploy; day 100.

X3. Data governance and client consent

What it is. A services firm holds client data because a person needed it to do the work. A platform holds it because the platform runs on it, across clients, with models processing it. The lever is the governance that lets a client say yes: what is processed where, by which models, retained how long, used for what (including whether it ever contributes to R10's data products), and disclosed how. The software edition's lever carried Strong evidence from regulators and standards on data governance and it carries here; the services-specific content is that the consent is renegotiated client by client at renewal (P4's calendar) and that a services client's trust was personal and is now institutional.

Line it moves. The multiple (a buyer prices governance it can audit); enables R10.

Grade. Strong (standards), carried.

How to run it. The data map (what is held, from whom, under what terms) is built in P8. The consent language goes into the engagement letter at renewal (O5). R10 waits until consent covers a majority of the book (P4's year-three metric).

Kill criteria. Any client data processed outside the terms the client agreed. Data-derivative consent below 50 percent of revenue at year three (R10 cannot be built).

Metrics. Year one: share of book on the new data terms. Year three: share consenting to derivative use; zero incidents.

Horizon and stage. Convert; day 100 to year one.

X4. Security posture

What it is. SOC 2, ISO 27001 or the sector's equivalent, because the client is now trusting a platform with its data rather than a person with a laptop, and because every enterprise client's procurement function will ask. Carried from the software edition with Medium evidence that posture moves win rate and multiple.⁵ For a services firm this is usually new: the firm had a confidentiality clause and a locked cabinet.

Line it moves. The multiple; win rate (R7).

Grade. Medium, carried.

How to run it. The platform company's certification may cover the delivery entity if the structure allows; check with P5's counsel. If not, start the audit in year one. Security questionnaires are an O1 automation.

Metrics. Year one: certification scope and status. Year three: zero incidents; questionnaire turnaround.

Horizon and stage. Deploy; year one.

X5. Key-person and partner retention

What it is. The people from P3's map, kept. Rollover equity so their upside is the exit; a role with a price on it (R11) so their work has a future; a compensation formula that pays for the conversion rather than against it (O7); and non-competes that are enforceable where the firm operates. This lever is the operational side of condition five and the single largest risk in the thesis, because the asset being bought is relationships held by people whose job is being automated.

Line it moves. Risk; book retention (R4).

Typical range and evidence. No independent evidence on partner departure after a change of ownership in professional services; the first study is in progress.⁶ A different population gives a sense of scale: an executive-search firm cites research finding 58 percent of PE-backed CEOs replaced within two years, which a relationship business cannot survive.⁷ Grade: Weak (no sector data); runs on the firm's own map and baseline.

How to run it. Structure (P7) does most of it before close. After close, the three instruments are O7, R11 and the visible senior role in D3's review model. Track named-person retention monthly, by name, at the board.

Kill criteria. Named-person retention below 90 percent in any quarter. Any relationship holder above 10 percent of revenue leaving (a book event, not a people event).

Metrics. Monthly: named-person retention. Year three: share of revenue held by people hired after close (P3's institutionalisation metric).

Horizon and stage. Convert; day 100 to year three.

X6. Exit readiness: the re-rating case

What it is. The scorecard, packaged for a buyer, with the honest alternative beside it. The case has three parts: what the firm is now (recurring share, gross margin, revenue per delivery FTE, NRR, error rate, all against the close baseline, audited), who buys it and how they price it (software strategics on a revenue multiple if the metrics clear their thresholds; larger sponsors on EBITDA with a premium for the conversion track record; conventional services buyers on EBITDA if neither), and what organic growth looks like once acquisitions are stripped out, because a buyer will strip them out.

Line it moves. The multiple.

Typical range and evidence. The comps are Strong and they are against the thesis: AI-invested outsourcers at 5 to 23x EV/EBITDA versus software peers at 22 to 92x, per the Fortune op-ed's comparison;¹ a public professional-services index at 6.1x EBITDA in mid-2026;² listed professional-services companies at a global median of 9.0x EV/EBITDA at the end of September 2026, 10.9x in the US.⁸ On the other side, the conversion vehicles that have raised capital have raised it at valuations that imply a software-like outcome (one at $12 billion in August 2026) with no audited margins published.⁹ The premium, if it exists, is unproven. Grade: Strong (comps) / Weak (premium).

How to run it. From year two, produce the exit scorecard quarterly as if a buyer were reading it: audited metrics, organic growth separated from acquired, the error rate and claims log, the data-consent share, the named-person retention. Identify the three buyer types and the metric thresholds each would need; the software strategic's thresholds are the Act II crossover thresholds in chapter 10. Run the model at all three multiples and show the board which case the firm is in. If it is in the services case at year three, say so and sell it as the best-run services firm in its segment, which is a real outcome and an honest one.

Kill criteria. None; this lever has no kill, only an answer.

Metrics. Year two onward: the exit scorecard; the implied multiple under each buyer case.

Horizon and stage. Compound; year two to exit.

The honest version of the exit

A conversion that answers all four of the critics' reasons on the scorecard has a software-like case to make, and should make it to software buyers. A conversion that answers two has made a services firm materially more profitable and more durable, and should be sold as one. The failure is not the second outcome; it is underwriting the first at entry and discovering the second at exit. P7 puts the two cases side by side in the model at the start so that the committee knows which it bought; X6 puts them side by side at the end so that the board knows which it is selling.